xref: /linux/drivers/nvme/host/fabrics.c (revision 16e5ac127d8d18adf85fe5ba847d77b58d1ed418)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVMe over Fabrics common host code.
4  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5  */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/init.h>
8 #include <linux/miscdevice.h>
9 #include <linux/module.h>
10 #include <linux/mutex.h>
11 #include <linux/parser.h>
12 #include <linux/seq_file.h>
13 #include "nvme.h"
14 #include "fabrics.h"
15 #include <linux/nvme-keyring.h>
16 
17 static LIST_HEAD(nvmf_transports);
18 static DECLARE_RWSEM(nvmf_transports_rwsem);
19 
20 static LIST_HEAD(nvmf_hosts);
21 static DEFINE_MUTEX(nvmf_hosts_mutex);
22 
23 static struct nvmf_host *nvmf_default_host;
24 
25 static struct nvmf_host *nvmf_host_alloc(const char *hostnqn, uuid_t *id)
26 {
27 	struct nvmf_host *host;
28 
29 	host = kmalloc(sizeof(*host), GFP_KERNEL);
30 	if (!host)
31 		return NULL;
32 
33 	kref_init(&host->ref);
34 	uuid_copy(&host->id, id);
35 	strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
36 
37 	return host;
38 }
39 
40 static struct nvmf_host *nvmf_host_add(const char *hostnqn, uuid_t *id)
41 {
42 	struct nvmf_host *host;
43 
44 	mutex_lock(&nvmf_hosts_mutex);
45 
46 	/*
47 	 * We have defined a host as how it is perceived by the target.
48 	 * Therefore, we don't allow different Host NQNs with the same Host ID.
49 	 * Similarly, we do not allow the usage of the same Host NQN with
50 	 * different Host IDs. This'll maintain unambiguous host identification.
51 	 */
52 	list_for_each_entry(host, &nvmf_hosts, list) {
53 		bool same_hostnqn = !strcmp(host->nqn, hostnqn);
54 		bool same_hostid = uuid_equal(&host->id, id);
55 
56 		if (same_hostnqn && same_hostid) {
57 			kref_get(&host->ref);
58 			goto out_unlock;
59 		}
60 		if (same_hostnqn) {
61 			pr_err("found same hostnqn %s but different hostid %pUb\n",
62 			       hostnqn, id);
63 			host = ERR_PTR(-EINVAL);
64 			goto out_unlock;
65 		}
66 		if (same_hostid) {
67 			pr_err("found same hostid %pUb but different hostnqn %s\n",
68 			       id, hostnqn);
69 			host = ERR_PTR(-EINVAL);
70 			goto out_unlock;
71 		}
72 	}
73 
74 	host = nvmf_host_alloc(hostnqn, id);
75 	if (!host) {
76 		host = ERR_PTR(-ENOMEM);
77 		goto out_unlock;
78 	}
79 
80 	list_add_tail(&host->list, &nvmf_hosts);
81 out_unlock:
82 	mutex_unlock(&nvmf_hosts_mutex);
83 	return host;
84 }
85 
86 static struct nvmf_host *nvmf_host_default(void)
87 {
88 	struct nvmf_host *host;
89 	char nqn[NVMF_NQN_SIZE];
90 	uuid_t id;
91 
92 	uuid_gen(&id);
93 	snprintf(nqn, NVMF_NQN_SIZE,
94 		"nqn.2014-08.org.nvmexpress:uuid:%pUb", &id);
95 
96 	host = nvmf_host_alloc(nqn, &id);
97 	if (!host)
98 		return NULL;
99 
100 	mutex_lock(&nvmf_hosts_mutex);
101 	list_add_tail(&host->list, &nvmf_hosts);
102 	mutex_unlock(&nvmf_hosts_mutex);
103 
104 	return host;
105 }
106 
107 static void nvmf_host_destroy(struct kref *ref)
108 {
109 	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
110 
111 	mutex_lock(&nvmf_hosts_mutex);
112 	list_del(&host->list);
113 	mutex_unlock(&nvmf_hosts_mutex);
114 
115 	kfree(host);
116 }
117 
118 static void nvmf_host_put(struct nvmf_host *host)
119 {
120 	if (host)
121 		kref_put(&host->ref, nvmf_host_destroy);
122 }
123 
124 /**
125  * nvmf_get_address() -  Get address/port
126  * @ctrl:	Host NVMe controller instance which we got the address
127  * @buf:	OUTPUT parameter that will contain the address/port
128  * @size:	buffer size
129  */
130 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
131 {
132 	int len = 0;
133 
134 	if (ctrl->opts->mask & NVMF_OPT_TRADDR)
135 		len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
136 	if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
137 		len += scnprintf(buf + len, size - len, "%strsvcid=%s",
138 				(len) ? "," : "", ctrl->opts->trsvcid);
139 	if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
140 		len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
141 				(len) ? "," : "", ctrl->opts->host_traddr);
142 	if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
143 		len += scnprintf(buf + len, size - len, "%shost_iface=%s",
144 				(len) ? "," : "", ctrl->opts->host_iface);
145 	len += scnprintf(buf + len, size - len, "\n");
146 
147 	return len;
148 }
149 EXPORT_SYMBOL_GPL(nvmf_get_address);
150 
151 /**
152  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
153  * @ctrl:	Host NVMe controller instance maintaining the admin
154  *		queue used to submit the property read command to
155  *		the allocated NVMe controller resource on the target system.
156  * @off:	Starting offset value of the targeted property
157  *		register (see the fabrics section of the NVMe standard).
158  * @val:	OUTPUT parameter that will contain the value of
159  *		the property after a successful read.
160  *
161  * Used by the host system to retrieve a 32-bit capsule property value
162  * from an NVMe controller on the target system.
163  *
164  * ("Capsule property" is an "PCIe register concept" applied to the
165  * NVMe fabrics space.)
166  *
167  * Return:
168  *	0: successful read
169  *	> 0: NVMe error status code
170  *	< 0: Linux errno error code
171  */
172 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
173 {
174 	struct nvme_command cmd = { };
175 	union nvme_result res;
176 	int ret;
177 
178 	cmd.prop_get.opcode = nvme_fabrics_command;
179 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
180 	cmd.prop_get.offset = cpu_to_le32(off);
181 
182 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
183 			NVME_QID_ANY, 0, 0);
184 
185 	if (ret >= 0)
186 		*val = le64_to_cpu(res.u64);
187 	if (unlikely(ret != 0))
188 		dev_err(ctrl->device,
189 			"Property Get error: %d, offset %#x\n",
190 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
191 
192 	return ret;
193 }
194 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
195 
196 /**
197  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
198  * @ctrl:	Host NVMe controller instance maintaining the admin
199  *		queue used to submit the property read command to
200  *		the allocated controller resource on the target system.
201  * @off:	Starting offset value of the targeted property
202  *		register (see the fabrics section of the NVMe standard).
203  * @val:	OUTPUT parameter that will contain the value of
204  *		the property after a successful read.
205  *
206  * Used by the host system to retrieve a 64-bit capsule property value
207  * from an NVMe controller on the target system.
208  *
209  * ("Capsule property" is an "PCIe register concept" applied to the
210  * NVMe fabrics space.)
211  *
212  * Return:
213  *	0: successful read
214  *	> 0: NVMe error status code
215  *	< 0: Linux errno error code
216  */
217 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
218 {
219 	struct nvme_command cmd = { };
220 	union nvme_result res;
221 	int ret;
222 
223 	cmd.prop_get.opcode = nvme_fabrics_command;
224 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
225 	cmd.prop_get.attrib = 1;
226 	cmd.prop_get.offset = cpu_to_le32(off);
227 
228 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
229 			NVME_QID_ANY, 0, 0);
230 
231 	if (ret >= 0)
232 		*val = le64_to_cpu(res.u64);
233 	if (unlikely(ret != 0))
234 		dev_err(ctrl->device,
235 			"Property Get error: %d, offset %#x\n",
236 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
237 	return ret;
238 }
239 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
240 
241 /**
242  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
243  * @ctrl:	Host NVMe controller instance maintaining the admin
244  *		queue used to submit the property read command to
245  *		the allocated NVMe controller resource on the target system.
246  * @off:	Starting offset value of the targeted property
247  *		register (see the fabrics section of the NVMe standard).
248  * @val:	Input parameter that contains the value to be
249  *		written to the property.
250  *
251  * Used by the NVMe host system to write a 32-bit capsule property value
252  * to an NVMe controller on the target system.
253  *
254  * ("Capsule property" is an "PCIe register concept" applied to the
255  * NVMe fabrics space.)
256  *
257  * Return:
258  *	0: successful write
259  *	> 0: NVMe error status code
260  *	< 0: Linux errno error code
261  */
262 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
263 {
264 	struct nvme_command cmd = { };
265 	int ret;
266 
267 	cmd.prop_set.opcode = nvme_fabrics_command;
268 	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
269 	cmd.prop_set.attrib = 0;
270 	cmd.prop_set.offset = cpu_to_le32(off);
271 	cmd.prop_set.value = cpu_to_le64(val);
272 
273 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0,
274 			NVME_QID_ANY, 0, 0);
275 	if (unlikely(ret))
276 		dev_err(ctrl->device,
277 			"Property Set error: %d, offset %#x\n",
278 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
279 	return ret;
280 }
281 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
282 
283 /**
284  * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
285  * 				connect() errors.
286  * @ctrl:	The specific /dev/nvmeX device that had the error.
287  * @errval:	Error code to be decoded in a more human-friendly
288  * 		printout.
289  * @offset:	For use with the NVMe error code
290  * 		NVME_SC_CONNECT_INVALID_PARAM.
291  * @cmd:	This is the SQE portion of a submission capsule.
292  * @data:	This is the "Data" portion of a submission capsule.
293  */
294 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
295 		int errval, int offset, struct nvme_command *cmd,
296 		struct nvmf_connect_data *data)
297 {
298 	int err_sctype = errval & ~NVME_SC_DNR;
299 
300 	if (errval < 0) {
301 		dev_err(ctrl->device,
302 			"Connect command failed, errno: %d\n", errval);
303 		return;
304 	}
305 
306 	switch (err_sctype) {
307 	case NVME_SC_CONNECT_INVALID_PARAM:
308 		if (offset >> 16) {
309 			char *inv_data = "Connect Invalid Data Parameter";
310 
311 			switch (offset & 0xffff) {
312 			case (offsetof(struct nvmf_connect_data, cntlid)):
313 				dev_err(ctrl->device,
314 					"%s, cntlid: %d\n",
315 					inv_data, data->cntlid);
316 				break;
317 			case (offsetof(struct nvmf_connect_data, hostnqn)):
318 				dev_err(ctrl->device,
319 					"%s, hostnqn \"%s\"\n",
320 					inv_data, data->hostnqn);
321 				break;
322 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
323 				dev_err(ctrl->device,
324 					"%s, subsysnqn \"%s\"\n",
325 					inv_data, data->subsysnqn);
326 				break;
327 			default:
328 				dev_err(ctrl->device,
329 					"%s, starting byte offset: %d\n",
330 				       inv_data, offset & 0xffff);
331 				break;
332 			}
333 		} else {
334 			char *inv_sqe = "Connect Invalid SQE Parameter";
335 
336 			switch (offset) {
337 			case (offsetof(struct nvmf_connect_command, qid)):
338 				dev_err(ctrl->device,
339 				       "%s, qid %d\n",
340 					inv_sqe, cmd->connect.qid);
341 				break;
342 			default:
343 				dev_err(ctrl->device,
344 					"%s, starting byte offset: %d\n",
345 					inv_sqe, offset);
346 			}
347 		}
348 		break;
349 	case NVME_SC_CONNECT_INVALID_HOST:
350 		dev_err(ctrl->device,
351 			"Connect for subsystem %s is not allowed, hostnqn: %s\n",
352 			data->subsysnqn, data->hostnqn);
353 		break;
354 	case NVME_SC_CONNECT_CTRL_BUSY:
355 		dev_err(ctrl->device,
356 			"Connect command failed: controller is busy or not available\n");
357 		break;
358 	case NVME_SC_CONNECT_FORMAT:
359 		dev_err(ctrl->device,
360 			"Connect incompatible format: %d",
361 			cmd->connect.recfmt);
362 		break;
363 	case NVME_SC_HOST_PATH_ERROR:
364 		dev_err(ctrl->device,
365 			"Connect command failed: host path error\n");
366 		break;
367 	case NVME_SC_AUTH_REQUIRED:
368 		dev_err(ctrl->device,
369 			"Connect command failed: authentication required\n");
370 		break;
371 	default:
372 		dev_err(ctrl->device,
373 			"Connect command failed, error wo/DNR bit: %d\n",
374 			err_sctype);
375 		break;
376 	}
377 }
378 
379 static struct nvmf_connect_data *nvmf_connect_data_prep(struct nvme_ctrl *ctrl,
380 		u16 cntlid)
381 {
382 	struct nvmf_connect_data *data;
383 
384 	data = kzalloc(sizeof(*data), GFP_KERNEL);
385 	if (!data)
386 		return NULL;
387 
388 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
389 	data->cntlid = cpu_to_le16(cntlid);
390 	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
391 	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
392 
393 	return data;
394 }
395 
396 static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,
397 		struct nvme_command *cmd)
398 {
399 	cmd->connect.opcode = nvme_fabrics_command;
400 	cmd->connect.fctype = nvme_fabrics_type_connect;
401 	cmd->connect.qid = cpu_to_le16(qid);
402 
403 	if (qid) {
404 		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
405 	} else {
406 		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
407 
408 		/*
409 		 * set keep-alive timeout in seconds granularity (ms * 1000)
410 		 */
411 		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
412 	}
413 
414 	if (ctrl->opts->disable_sqflow)
415 		cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
416 }
417 
418 /**
419  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
420  *				API function.
421  * @ctrl:	Host nvme controller instance used to request
422  *              a new NVMe controller allocation on the target
423  *              system and  establish an NVMe Admin connection to
424  *              that controller.
425  *
426  * This function enables an NVMe host device to request a new allocation of
427  * an NVMe controller resource on a target system as well establish a
428  * fabrics-protocol connection of the NVMe Admin queue between the
429  * host system device and the allocated NVMe controller on the
430  * target system via a NVMe Fabrics "Connect" command.
431  *
432  * Return:
433  *	0: success
434  *	> 0: NVMe error status code
435  *	< 0: Linux errno error code
436  *
437  */
438 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
439 {
440 	struct nvme_command cmd = { };
441 	union nvme_result res;
442 	struct nvmf_connect_data *data;
443 	int ret;
444 	u32 result;
445 
446 	nvmf_connect_cmd_prep(ctrl, 0, &cmd);
447 
448 	data = nvmf_connect_data_prep(ctrl, 0xffff);
449 	if (!data)
450 		return -ENOMEM;
451 
452 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
453 			data, sizeof(*data), NVME_QID_ANY, 1,
454 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
455 	if (ret) {
456 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
457 				       &cmd, data);
458 		goto out_free_data;
459 	}
460 
461 	result = le32_to_cpu(res.u32);
462 	ctrl->cntlid = result & 0xFFFF;
463 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
464 		/* Secure concatenation is not implemented */
465 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
466 			dev_warn(ctrl->device,
467 				 "qid 0: secure concatenation is not supported\n");
468 			ret = NVME_SC_AUTH_REQUIRED;
469 			goto out_free_data;
470 		}
471 		/* Authentication required */
472 		ret = nvme_auth_negotiate(ctrl, 0);
473 		if (ret) {
474 			dev_warn(ctrl->device,
475 				 "qid 0: authentication setup failed\n");
476 			ret = NVME_SC_AUTH_REQUIRED;
477 			goto out_free_data;
478 		}
479 		ret = nvme_auth_wait(ctrl, 0);
480 		if (ret)
481 			dev_warn(ctrl->device,
482 				 "qid 0: authentication failed\n");
483 		else
484 			dev_info(ctrl->device,
485 				 "qid 0: authenticated\n");
486 	}
487 out_free_data:
488 	kfree(data);
489 	return ret;
490 }
491 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
492 
493 /**
494  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
495  *			     API function.
496  * @ctrl:	Host nvme controller instance used to establish an
497  *		NVMe I/O queue connection to the already allocated NVMe
498  *		controller on the target system.
499  * @qid:	NVMe I/O queue number for the new I/O connection between
500  *		host and target (note qid == 0 is illegal as this is
501  *		the Admin queue, per NVMe standard).
502  *
503  * This function issues a fabrics-protocol connection
504  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
505  * between the host system device and the allocated NVMe controller
506  * on the target system.
507  *
508  * Return:
509  *	0: success
510  *	> 0: NVMe error status code
511  *	< 0: Linux errno error code
512  */
513 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
514 {
515 	struct nvme_command cmd = { };
516 	struct nvmf_connect_data *data;
517 	union nvme_result res;
518 	int ret;
519 	u32 result;
520 
521 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
522 
523 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
524 	if (!data)
525 		return -ENOMEM;
526 
527 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
528 			data, sizeof(*data), qid, 1,
529 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
530 	if (ret) {
531 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
532 				       &cmd, data);
533 	}
534 	result = le32_to_cpu(res.u32);
535 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
536 		/* Secure concatenation is not implemented */
537 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
538 			dev_warn(ctrl->device,
539 				 "qid 0: secure concatenation is not supported\n");
540 			ret = NVME_SC_AUTH_REQUIRED;
541 			goto out_free_data;
542 		}
543 		/* Authentication required */
544 		ret = nvme_auth_negotiate(ctrl, qid);
545 		if (ret) {
546 			dev_warn(ctrl->device,
547 				 "qid %d: authentication setup failed\n", qid);
548 			ret = NVME_SC_AUTH_REQUIRED;
549 		} else {
550 			ret = nvme_auth_wait(ctrl, qid);
551 			if (ret)
552 				dev_warn(ctrl->device,
553 					 "qid %u: authentication failed\n", qid);
554 		}
555 	}
556 out_free_data:
557 	kfree(data);
558 	return ret;
559 }
560 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
561 
562 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
563 {
564 	if (ctrl->opts->max_reconnects == -1 ||
565 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
566 		return true;
567 
568 	return false;
569 }
570 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
571 
572 /**
573  * nvmf_register_transport() - NVMe Fabrics Library registration function.
574  * @ops:	Transport ops instance to be registered to the
575  *		common fabrics library.
576  *
577  * API function that registers the type of specific transport fabric
578  * being implemented to the common NVMe fabrics library. Part of
579  * the overall init sequence of starting up a fabrics driver.
580  */
581 int nvmf_register_transport(struct nvmf_transport_ops *ops)
582 {
583 	if (!ops->create_ctrl)
584 		return -EINVAL;
585 
586 	down_write(&nvmf_transports_rwsem);
587 	list_add_tail(&ops->entry, &nvmf_transports);
588 	up_write(&nvmf_transports_rwsem);
589 
590 	return 0;
591 }
592 EXPORT_SYMBOL_GPL(nvmf_register_transport);
593 
594 /**
595  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
596  * @ops:	Transport ops instance to be unregistered from the
597  *		common fabrics library.
598  *
599  * Fabrics API function that unregisters the type of specific transport
600  * fabric being implemented from the common NVMe fabrics library.
601  * Part of the overall exit sequence of unloading the implemented driver.
602  */
603 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
604 {
605 	down_write(&nvmf_transports_rwsem);
606 	list_del(&ops->entry);
607 	up_write(&nvmf_transports_rwsem);
608 }
609 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
610 
611 static struct nvmf_transport_ops *nvmf_lookup_transport(
612 		struct nvmf_ctrl_options *opts)
613 {
614 	struct nvmf_transport_ops *ops;
615 
616 	lockdep_assert_held(&nvmf_transports_rwsem);
617 
618 	list_for_each_entry(ops, &nvmf_transports, entry) {
619 		if (strcmp(ops->name, opts->transport) == 0)
620 			return ops;
621 	}
622 
623 	return NULL;
624 }
625 
626 static struct key *nvmf_parse_key(int key_id)
627 {
628 	struct key *key;
629 
630 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
631 		pr_err("TLS is not supported\n");
632 		return ERR_PTR(-EINVAL);
633 	}
634 
635 	key = key_lookup(key_id);
636 	if (!IS_ERR(key))
637 		pr_err("key id %08x not found\n", key_id);
638 	else
639 		pr_debug("Using key id %08x\n", key_id);
640 	return key;
641 }
642 
643 static const match_table_t opt_tokens = {
644 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
645 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
646 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
647 	{ NVMF_OPT_NQN,			"nqn=%s"		},
648 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
649 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
650 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
651 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
652 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
653 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
654 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
655 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
656 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
657 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
658 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
659 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
660 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
661 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
662 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
663 	{ NVMF_OPT_TOS,			"tos=%d"		},
664 #ifdef CONFIG_NVME_TCP_TLS
665 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
666 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
667 #endif
668 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
669 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
670 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
671 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
672 #ifdef CONFIG_NVME_TCP_TLS
673 	{ NVMF_OPT_TLS,			"tls"			},
674 #endif
675 	{ NVMF_OPT_ERR,			NULL			}
676 };
677 
678 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
679 		const char *buf)
680 {
681 	substring_t args[MAX_OPT_ARGS];
682 	char *options, *o, *p;
683 	int token, ret = 0;
684 	size_t nqnlen  = 0;
685 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
686 	uuid_t hostid;
687 	char hostnqn[NVMF_NQN_SIZE];
688 	struct key *key;
689 
690 	/* Set defaults */
691 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
692 	opts->nr_io_queues = num_online_cpus();
693 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
694 	opts->kato = 0;
695 	opts->duplicate_connect = false;
696 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
697 	opts->hdr_digest = false;
698 	opts->data_digest = false;
699 	opts->tos = -1; /* < 0 == use transport default */
700 	opts->tls = false;
701 	opts->tls_key = NULL;
702 	opts->keyring = NULL;
703 
704 	options = o = kstrdup(buf, GFP_KERNEL);
705 	if (!options)
706 		return -ENOMEM;
707 
708 	/* use default host if not given by user space */
709 	uuid_copy(&hostid, &nvmf_default_host->id);
710 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
711 
712 	while ((p = strsep(&o, ",\n")) != NULL) {
713 		if (!*p)
714 			continue;
715 
716 		token = match_token(p, opt_tokens, args);
717 		opts->mask |= token;
718 		switch (token) {
719 		case NVMF_OPT_TRANSPORT:
720 			p = match_strdup(args);
721 			if (!p) {
722 				ret = -ENOMEM;
723 				goto out;
724 			}
725 			kfree(opts->transport);
726 			opts->transport = p;
727 			break;
728 		case NVMF_OPT_NQN:
729 			p = match_strdup(args);
730 			if (!p) {
731 				ret = -ENOMEM;
732 				goto out;
733 			}
734 			kfree(opts->subsysnqn);
735 			opts->subsysnqn = p;
736 			nqnlen = strlen(opts->subsysnqn);
737 			if (nqnlen >= NVMF_NQN_SIZE) {
738 				pr_err("%s needs to be < %d bytes\n",
739 					opts->subsysnqn, NVMF_NQN_SIZE);
740 				ret = -EINVAL;
741 				goto out;
742 			}
743 			opts->discovery_nqn =
744 				!(strcmp(opts->subsysnqn,
745 					 NVME_DISC_SUBSYS_NAME));
746 			break;
747 		case NVMF_OPT_TRADDR:
748 			p = match_strdup(args);
749 			if (!p) {
750 				ret = -ENOMEM;
751 				goto out;
752 			}
753 			kfree(opts->traddr);
754 			opts->traddr = p;
755 			break;
756 		case NVMF_OPT_TRSVCID:
757 			p = match_strdup(args);
758 			if (!p) {
759 				ret = -ENOMEM;
760 				goto out;
761 			}
762 			kfree(opts->trsvcid);
763 			opts->trsvcid = p;
764 			break;
765 		case NVMF_OPT_QUEUE_SIZE:
766 			if (match_int(args, &token)) {
767 				ret = -EINVAL;
768 				goto out;
769 			}
770 			if (token < NVMF_MIN_QUEUE_SIZE ||
771 			    token > NVMF_MAX_QUEUE_SIZE) {
772 				pr_err("Invalid queue_size %d\n", token);
773 				ret = -EINVAL;
774 				goto out;
775 			}
776 			opts->queue_size = token;
777 			break;
778 		case NVMF_OPT_NR_IO_QUEUES:
779 			if (match_int(args, &token)) {
780 				ret = -EINVAL;
781 				goto out;
782 			}
783 			if (token <= 0) {
784 				pr_err("Invalid number of IOQs %d\n", token);
785 				ret = -EINVAL;
786 				goto out;
787 			}
788 			if (opts->discovery_nqn) {
789 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
790 				break;
791 			}
792 
793 			opts->nr_io_queues = min_t(unsigned int,
794 					num_online_cpus(), token);
795 			break;
796 		case NVMF_OPT_KATO:
797 			if (match_int(args, &token)) {
798 				ret = -EINVAL;
799 				goto out;
800 			}
801 
802 			if (token < 0) {
803 				pr_err("Invalid keep_alive_tmo %d\n", token);
804 				ret = -EINVAL;
805 				goto out;
806 			} else if (token == 0 && !opts->discovery_nqn) {
807 				/* Allowed for debug */
808 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
809 			}
810 			opts->kato = token;
811 			break;
812 		case NVMF_OPT_CTRL_LOSS_TMO:
813 			if (match_int(args, &token)) {
814 				ret = -EINVAL;
815 				goto out;
816 			}
817 
818 			if (token < 0)
819 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
820 			ctrl_loss_tmo = token;
821 			break;
822 		case NVMF_OPT_FAIL_FAST_TMO:
823 			if (match_int(args, &token)) {
824 				ret = -EINVAL;
825 				goto out;
826 			}
827 
828 			if (token >= 0)
829 				pr_warn("I/O fail on reconnect controller after %d sec\n",
830 					token);
831 			else
832 				token = -1;
833 
834 			opts->fast_io_fail_tmo = token;
835 			break;
836 		case NVMF_OPT_HOSTNQN:
837 			if (opts->host) {
838 				pr_err("hostnqn already user-assigned: %s\n",
839 				       opts->host->nqn);
840 				ret = -EADDRINUSE;
841 				goto out;
842 			}
843 			p = match_strdup(args);
844 			if (!p) {
845 				ret = -ENOMEM;
846 				goto out;
847 			}
848 			nqnlen = strlen(p);
849 			if (nqnlen >= NVMF_NQN_SIZE) {
850 				pr_err("%s needs to be < %d bytes\n",
851 					p, NVMF_NQN_SIZE);
852 				kfree(p);
853 				ret = -EINVAL;
854 				goto out;
855 			}
856 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
857 			kfree(p);
858 			break;
859 		case NVMF_OPT_RECONNECT_DELAY:
860 			if (match_int(args, &token)) {
861 				ret = -EINVAL;
862 				goto out;
863 			}
864 			if (token <= 0) {
865 				pr_err("Invalid reconnect_delay %d\n", token);
866 				ret = -EINVAL;
867 				goto out;
868 			}
869 			opts->reconnect_delay = token;
870 			break;
871 		case NVMF_OPT_HOST_TRADDR:
872 			p = match_strdup(args);
873 			if (!p) {
874 				ret = -ENOMEM;
875 				goto out;
876 			}
877 			kfree(opts->host_traddr);
878 			opts->host_traddr = p;
879 			break;
880 		case NVMF_OPT_HOST_IFACE:
881 			p = match_strdup(args);
882 			if (!p) {
883 				ret = -ENOMEM;
884 				goto out;
885 			}
886 			kfree(opts->host_iface);
887 			opts->host_iface = p;
888 			break;
889 		case NVMF_OPT_HOST_ID:
890 			p = match_strdup(args);
891 			if (!p) {
892 				ret = -ENOMEM;
893 				goto out;
894 			}
895 			ret = uuid_parse(p, &hostid);
896 			if (ret) {
897 				pr_err("Invalid hostid %s\n", p);
898 				ret = -EINVAL;
899 				kfree(p);
900 				goto out;
901 			}
902 			kfree(p);
903 			break;
904 		case NVMF_OPT_DUP_CONNECT:
905 			opts->duplicate_connect = true;
906 			break;
907 		case NVMF_OPT_DISABLE_SQFLOW:
908 			opts->disable_sqflow = true;
909 			break;
910 		case NVMF_OPT_HDR_DIGEST:
911 			opts->hdr_digest = true;
912 			break;
913 		case NVMF_OPT_DATA_DIGEST:
914 			opts->data_digest = true;
915 			break;
916 		case NVMF_OPT_NR_WRITE_QUEUES:
917 			if (match_int(args, &token)) {
918 				ret = -EINVAL;
919 				goto out;
920 			}
921 			if (token <= 0) {
922 				pr_err("Invalid nr_write_queues %d\n", token);
923 				ret = -EINVAL;
924 				goto out;
925 			}
926 			opts->nr_write_queues = token;
927 			break;
928 		case NVMF_OPT_NR_POLL_QUEUES:
929 			if (match_int(args, &token)) {
930 				ret = -EINVAL;
931 				goto out;
932 			}
933 			if (token <= 0) {
934 				pr_err("Invalid nr_poll_queues %d\n", token);
935 				ret = -EINVAL;
936 				goto out;
937 			}
938 			opts->nr_poll_queues = token;
939 			break;
940 		case NVMF_OPT_TOS:
941 			if (match_int(args, &token)) {
942 				ret = -EINVAL;
943 				goto out;
944 			}
945 			if (token < 0) {
946 				pr_err("Invalid type of service %d\n", token);
947 				ret = -EINVAL;
948 				goto out;
949 			}
950 			if (token > 255) {
951 				pr_warn("Clamping type of service to 255\n");
952 				token = 255;
953 			}
954 			opts->tos = token;
955 			break;
956 		case NVMF_OPT_KEYRING:
957 			if (match_int(args, &key_id) || key_id <= 0) {
958 				ret = -EINVAL;
959 				goto out;
960 			}
961 			key = nvmf_parse_key(key_id);
962 			if (IS_ERR(key)) {
963 				ret = PTR_ERR(key);
964 				goto out;
965 			}
966 			key_put(opts->keyring);
967 			opts->keyring = key;
968 			break;
969 		case NVMF_OPT_TLS_KEY:
970 			if (match_int(args, &key_id) || key_id <= 0) {
971 				ret = -EINVAL;
972 				goto out;
973 			}
974 			key = nvmf_parse_key(key_id);
975 			if (IS_ERR(key)) {
976 				ret = PTR_ERR(key);
977 				goto out;
978 			}
979 			key_put(opts->tls_key);
980 			opts->tls_key = key;
981 			break;
982 		case NVMF_OPT_DISCOVERY:
983 			opts->discovery_nqn = true;
984 			break;
985 		case NVMF_OPT_DHCHAP_SECRET:
986 			p = match_strdup(args);
987 			if (!p) {
988 				ret = -ENOMEM;
989 				goto out;
990 			}
991 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
992 				pr_err("Invalid DH-CHAP secret %s\n", p);
993 				ret = -EINVAL;
994 				goto out;
995 			}
996 			kfree(opts->dhchap_secret);
997 			opts->dhchap_secret = p;
998 			break;
999 		case NVMF_OPT_DHCHAP_CTRL_SECRET:
1000 			p = match_strdup(args);
1001 			if (!p) {
1002 				ret = -ENOMEM;
1003 				goto out;
1004 			}
1005 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1006 				pr_err("Invalid DH-CHAP secret %s\n", p);
1007 				ret = -EINVAL;
1008 				goto out;
1009 			}
1010 			kfree(opts->dhchap_ctrl_secret);
1011 			opts->dhchap_ctrl_secret = p;
1012 			break;
1013 		case NVMF_OPT_TLS:
1014 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1015 				pr_err("TLS is not supported\n");
1016 				ret = -EINVAL;
1017 				goto out;
1018 			}
1019 			opts->tls = true;
1020 			break;
1021 		default:
1022 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1023 				p);
1024 			ret = -EINVAL;
1025 			goto out;
1026 		}
1027 	}
1028 
1029 	if (opts->discovery_nqn) {
1030 		opts->nr_io_queues = 0;
1031 		opts->nr_write_queues = 0;
1032 		opts->nr_poll_queues = 0;
1033 		opts->duplicate_connect = true;
1034 	} else {
1035 		if (!opts->kato)
1036 			opts->kato = NVME_DEFAULT_KATO;
1037 	}
1038 	if (ctrl_loss_tmo < 0) {
1039 		opts->max_reconnects = -1;
1040 	} else {
1041 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1042 						opts->reconnect_delay);
1043 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1044 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1045 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1046 	}
1047 
1048 	opts->host = nvmf_host_add(hostnqn, &hostid);
1049 	if (IS_ERR(opts->host)) {
1050 		ret = PTR_ERR(opts->host);
1051 		opts->host = NULL;
1052 		goto out;
1053 	}
1054 
1055 out:
1056 	kfree(options);
1057 	return ret;
1058 }
1059 
1060 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1061 			u32 io_queues[HCTX_MAX_TYPES])
1062 {
1063 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1064 		/*
1065 		 * separate read/write queues
1066 		 * hand out dedicated default queues only after we have
1067 		 * sufficient read queues.
1068 		 */
1069 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1070 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1071 		io_queues[HCTX_TYPE_DEFAULT] =
1072 			min(opts->nr_write_queues, nr_io_queues);
1073 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1074 	} else {
1075 		/*
1076 		 * shared read/write queues
1077 		 * either no write queues were requested, or we don't have
1078 		 * sufficient queue count to have dedicated default queues.
1079 		 */
1080 		io_queues[HCTX_TYPE_DEFAULT] =
1081 			min(opts->nr_io_queues, nr_io_queues);
1082 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1083 	}
1084 
1085 	if (opts->nr_poll_queues && nr_io_queues) {
1086 		/* map dedicated poll queues only if we have queues left */
1087 		io_queues[HCTX_TYPE_POLL] =
1088 			min(opts->nr_poll_queues, nr_io_queues);
1089 	}
1090 }
1091 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1092 
1093 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1094 		     u32 io_queues[HCTX_MAX_TYPES])
1095 {
1096 	struct nvmf_ctrl_options *opts = ctrl->opts;
1097 
1098 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1099 		/* separate read/write queues */
1100 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1101 			io_queues[HCTX_TYPE_DEFAULT];
1102 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1103 		set->map[HCTX_TYPE_READ].nr_queues =
1104 			io_queues[HCTX_TYPE_READ];
1105 		set->map[HCTX_TYPE_READ].queue_offset =
1106 			io_queues[HCTX_TYPE_DEFAULT];
1107 	} else {
1108 		/* shared read/write queues */
1109 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1110 			io_queues[HCTX_TYPE_DEFAULT];
1111 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1112 		set->map[HCTX_TYPE_READ].nr_queues =
1113 			io_queues[HCTX_TYPE_DEFAULT];
1114 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1115 	}
1116 
1117 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1118 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1119 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1120 		/* map dedicated poll queues only if we have queues left */
1121 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1122 		set->map[HCTX_TYPE_POLL].queue_offset =
1123 			io_queues[HCTX_TYPE_DEFAULT] +
1124 			io_queues[HCTX_TYPE_READ];
1125 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1126 	}
1127 
1128 	dev_info(ctrl->device,
1129 		"mapped %d/%d/%d default/read/poll queues.\n",
1130 		io_queues[HCTX_TYPE_DEFAULT],
1131 		io_queues[HCTX_TYPE_READ],
1132 		io_queues[HCTX_TYPE_POLL]);
1133 }
1134 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1135 
1136 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1137 		unsigned int required_opts)
1138 {
1139 	if ((opts->mask & required_opts) != required_opts) {
1140 		unsigned int i;
1141 
1142 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1143 			if ((opt_tokens[i].token & required_opts) &&
1144 			    !(opt_tokens[i].token & opts->mask)) {
1145 				pr_warn("missing parameter '%s'\n",
1146 					opt_tokens[i].pattern);
1147 			}
1148 		}
1149 
1150 		return -EINVAL;
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1157 		struct nvmf_ctrl_options *opts)
1158 {
1159 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1160 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1161 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1162 		return false;
1163 
1164 	/*
1165 	 * Checking the local address or host interfaces is rough.
1166 	 *
1167 	 * In most cases, none is specified and the host port or
1168 	 * host interface is selected by the stack.
1169 	 *
1170 	 * Assume no match if:
1171 	 * -  local address or host interface is specified and address
1172 	 *    or host interface is not the same
1173 	 * -  local address or host interface is not specified but
1174 	 *    remote is, or vice versa (admin using specific
1175 	 *    host_traddr/host_iface when it matters).
1176 	 */
1177 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1178 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1179 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1180 			return false;
1181 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1182 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1183 		return false;
1184 	}
1185 
1186 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1187 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1188 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1189 			return false;
1190 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1191 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1192 		return false;
1193 	}
1194 
1195 	return true;
1196 }
1197 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1198 
1199 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1200 		unsigned int allowed_opts)
1201 {
1202 	if (opts->mask & ~allowed_opts) {
1203 		unsigned int i;
1204 
1205 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1206 			if ((opt_tokens[i].token & opts->mask) &&
1207 			    (opt_tokens[i].token & ~allowed_opts)) {
1208 				pr_warn("invalid parameter '%s'\n",
1209 					opt_tokens[i].pattern);
1210 			}
1211 		}
1212 
1213 		return -EINVAL;
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1220 {
1221 	nvmf_host_put(opts->host);
1222 	key_put(opts->keyring);
1223 	key_put(opts->tls_key);
1224 	kfree(opts->transport);
1225 	kfree(opts->traddr);
1226 	kfree(opts->trsvcid);
1227 	kfree(opts->subsysnqn);
1228 	kfree(opts->host_traddr);
1229 	kfree(opts->host_iface);
1230 	kfree(opts->dhchap_secret);
1231 	kfree(opts->dhchap_ctrl_secret);
1232 	kfree(opts);
1233 }
1234 EXPORT_SYMBOL_GPL(nvmf_free_options);
1235 
1236 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1237 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1238 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1239 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1240 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1241 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1242 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1243 
1244 static struct nvme_ctrl *
1245 nvmf_create_ctrl(struct device *dev, const char *buf)
1246 {
1247 	struct nvmf_ctrl_options *opts;
1248 	struct nvmf_transport_ops *ops;
1249 	struct nvme_ctrl *ctrl;
1250 	int ret;
1251 
1252 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1253 	if (!opts)
1254 		return ERR_PTR(-ENOMEM);
1255 
1256 	ret = nvmf_parse_options(opts, buf);
1257 	if (ret)
1258 		goto out_free_opts;
1259 
1260 
1261 	request_module("nvme-%s", opts->transport);
1262 
1263 	/*
1264 	 * Check the generic options first as we need a valid transport for
1265 	 * the lookup below.  Then clear the generic flags so that transport
1266 	 * drivers don't have to care about them.
1267 	 */
1268 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1269 	if (ret)
1270 		goto out_free_opts;
1271 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1272 
1273 	down_read(&nvmf_transports_rwsem);
1274 	ops = nvmf_lookup_transport(opts);
1275 	if (!ops) {
1276 		pr_info("no handler found for transport %s.\n",
1277 			opts->transport);
1278 		ret = -EINVAL;
1279 		goto out_unlock;
1280 	}
1281 
1282 	if (!try_module_get(ops->module)) {
1283 		ret = -EBUSY;
1284 		goto out_unlock;
1285 	}
1286 	up_read(&nvmf_transports_rwsem);
1287 
1288 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1289 	if (ret)
1290 		goto out_module_put;
1291 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1292 				ops->allowed_opts | ops->required_opts);
1293 	if (ret)
1294 		goto out_module_put;
1295 
1296 	ctrl = ops->create_ctrl(dev, opts);
1297 	if (IS_ERR(ctrl)) {
1298 		ret = PTR_ERR(ctrl);
1299 		goto out_module_put;
1300 	}
1301 
1302 	module_put(ops->module);
1303 	return ctrl;
1304 
1305 out_module_put:
1306 	module_put(ops->module);
1307 	goto out_free_opts;
1308 out_unlock:
1309 	up_read(&nvmf_transports_rwsem);
1310 out_free_opts:
1311 	nvmf_free_options(opts);
1312 	return ERR_PTR(ret);
1313 }
1314 
1315 static struct class *nvmf_class;
1316 static struct device *nvmf_device;
1317 static DEFINE_MUTEX(nvmf_dev_mutex);
1318 
1319 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1320 		size_t count, loff_t *pos)
1321 {
1322 	struct seq_file *seq_file = file->private_data;
1323 	struct nvme_ctrl *ctrl;
1324 	const char *buf;
1325 	int ret = 0;
1326 
1327 	if (count > PAGE_SIZE)
1328 		return -ENOMEM;
1329 
1330 	buf = memdup_user_nul(ubuf, count);
1331 	if (IS_ERR(buf))
1332 		return PTR_ERR(buf);
1333 
1334 	mutex_lock(&nvmf_dev_mutex);
1335 	if (seq_file->private) {
1336 		ret = -EINVAL;
1337 		goto out_unlock;
1338 	}
1339 
1340 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1341 	if (IS_ERR(ctrl)) {
1342 		ret = PTR_ERR(ctrl);
1343 		goto out_unlock;
1344 	}
1345 
1346 	seq_file->private = ctrl;
1347 
1348 out_unlock:
1349 	mutex_unlock(&nvmf_dev_mutex);
1350 	kfree(buf);
1351 	return ret ? ret : count;
1352 }
1353 
1354 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1355 {
1356 	const struct match_token *tok;
1357 	int idx;
1358 
1359 	/*
1360 	 * Add dummy entries for instance and cntlid to
1361 	 * signal an invalid/non-existing controller
1362 	 */
1363 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1364 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1365 		tok = &opt_tokens[idx];
1366 		if (tok->token == NVMF_OPT_ERR)
1367 			continue;
1368 		seq_puts(seq_file, ",");
1369 		seq_puts(seq_file, tok->pattern);
1370 	}
1371 	seq_puts(seq_file, "\n");
1372 }
1373 
1374 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1375 {
1376 	struct nvme_ctrl *ctrl;
1377 
1378 	mutex_lock(&nvmf_dev_mutex);
1379 	ctrl = seq_file->private;
1380 	if (!ctrl) {
1381 		__nvmf_concat_opt_tokens(seq_file);
1382 		goto out_unlock;
1383 	}
1384 
1385 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1386 			ctrl->instance, ctrl->cntlid);
1387 
1388 out_unlock:
1389 	mutex_unlock(&nvmf_dev_mutex);
1390 	return 0;
1391 }
1392 
1393 static int nvmf_dev_open(struct inode *inode, struct file *file)
1394 {
1395 	/*
1396 	 * The miscdevice code initializes file->private_data, but doesn't
1397 	 * make use of it later.
1398 	 */
1399 	file->private_data = NULL;
1400 	return single_open(file, nvmf_dev_show, NULL);
1401 }
1402 
1403 static int nvmf_dev_release(struct inode *inode, struct file *file)
1404 {
1405 	struct seq_file *seq_file = file->private_data;
1406 	struct nvme_ctrl *ctrl = seq_file->private;
1407 
1408 	if (ctrl)
1409 		nvme_put_ctrl(ctrl);
1410 	return single_release(inode, file);
1411 }
1412 
1413 static const struct file_operations nvmf_dev_fops = {
1414 	.owner		= THIS_MODULE,
1415 	.write		= nvmf_dev_write,
1416 	.read		= seq_read,
1417 	.open		= nvmf_dev_open,
1418 	.release	= nvmf_dev_release,
1419 };
1420 
1421 static struct miscdevice nvmf_misc = {
1422 	.minor		= MISC_DYNAMIC_MINOR,
1423 	.name           = "nvme-fabrics",
1424 	.fops		= &nvmf_dev_fops,
1425 };
1426 
1427 static int __init nvmf_init(void)
1428 {
1429 	int ret;
1430 
1431 	nvmf_default_host = nvmf_host_default();
1432 	if (!nvmf_default_host)
1433 		return -ENOMEM;
1434 
1435 	nvmf_class = class_create("nvme-fabrics");
1436 	if (IS_ERR(nvmf_class)) {
1437 		pr_err("couldn't register class nvme-fabrics\n");
1438 		ret = PTR_ERR(nvmf_class);
1439 		goto out_free_host;
1440 	}
1441 
1442 	nvmf_device =
1443 		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1444 	if (IS_ERR(nvmf_device)) {
1445 		pr_err("couldn't create nvme-fabrics device!\n");
1446 		ret = PTR_ERR(nvmf_device);
1447 		goto out_destroy_class;
1448 	}
1449 
1450 	ret = misc_register(&nvmf_misc);
1451 	if (ret) {
1452 		pr_err("couldn't register misc device: %d\n", ret);
1453 		goto out_destroy_device;
1454 	}
1455 
1456 	return 0;
1457 
1458 out_destroy_device:
1459 	device_destroy(nvmf_class, MKDEV(0, 0));
1460 out_destroy_class:
1461 	class_destroy(nvmf_class);
1462 out_free_host:
1463 	nvmf_host_put(nvmf_default_host);
1464 	return ret;
1465 }
1466 
1467 static void __exit nvmf_exit(void)
1468 {
1469 	misc_deregister(&nvmf_misc);
1470 	device_destroy(nvmf_class, MKDEV(0, 0));
1471 	class_destroy(nvmf_class);
1472 	nvmf_host_put(nvmf_default_host);
1473 
1474 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1475 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1476 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1477 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1478 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1479 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1480 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1481 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1482 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1483 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1484 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1485 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1486 }
1487 
1488 MODULE_LICENSE("GPL v2");
1489 
1490 module_init(nvmf_init);
1491 module_exit(nvmf_exit);
1492